v1.0.0 · MIT · local-first

Is your repo ready for AI coding agents?

agentlint scans any codebase and gives it a 0–100 readiness score for Claude Code, Cursor, Codex, Copilot, and Gemini CLI. Local. Free. 30 seconds.

How it scores →Why this exists →GitHub →

~/your-repo

$ npx @agentlinthq/cli .

agentlint v1.0.0  scanning ~/your-repo

  Discoverability....... 25/25
  Buildability.......... 19/25
  Convention clarity.... 20/20
  Documentation surface. -/-   (skipped)
  Safety & guardrails... 12/15

  Overall ............... 87/100

  21 passes, 3 fails, 0 warnings, 6 skipped.
  Full report:    agentlint-report.html

What it checks

Five categories. Thirty checks. One score.

Discoverability

≤ 25 pts

Does an agent dropped into this repo know what it is, what it builds, and where the entry point lives? README, AGENTS.md, repo metadata, setup signals.

Buildability

≤ 25 pts

Can the agent actually run the project without asking? Lockfiles, declared scripts, env templates, working test commands.

Conventions, docs, safety

≤ 50 pts

Are commits conventional? Is there a license, a code of conduct, a security policy? Are there secrets in the tree?

Why this exists

Coding agents fail silently when they land in a repo they can't understand. They guess. They hallucinate paths. They make up commands. The fix is not a smarter agent — it's a clearer repo.

agentlint is the Lighthouse for that problem. It runs against your repo, gives you a 0–100 score, tells you exactly which signals are missing, and links to the fix. The CLI is local-first and stays free forever. The hosted dashboard is how the project sustains itself.

How it scores

Five categories, renormalized to 100.

Category	Covers	Weight
Discoverability	README, AGENTS.md, repo metadata	25
Buildability	Lockfile, scripts, env template	25
Convention clarity	Commit format, branch hygiene	20
Documentation surface	License, contributing, code of conduct	15
Safety & guardrails	Secrets, SECURITY.md, dependency hygiene	15

Score is renormalized against applicable categories. Skipped categories don't penalize the score. Full formula → ADR-0003

How it's built

agentlint is operated by an agent.

An autonomous coding agent (Claude Code) ships every line of this project. A human approves releases. The constitution that governs the agent is public:

CHARTER.md — what the agent decides alone, what it confirms, what it escalates.
PROJECT_STATE.md — live snapshot of what's shipped and what's next.
DECISIONS.md — the full ADR log.

The bet: a project run on agent-readable infrastructure is the most credible advocate for agent-readable infrastructure.

Hosted dashboard

Run history, PR comments, team badges.

The CLI is free forever. The dashboard is opt-in. Nothing in agentlint phones home by default.

See pricing →